I'm a software engineer from Gujarat, India, finishing a Computer Engineering degree β and building products that are meant to be deployed, not demoed. My work sits where AI systems meet product engineering: Python/FastAPI services, React/Next.js interfaces, Postgres data models, and LLM pipelines that have to behave under real load.
The two projects I'm proudest of take opposite approaches on purpose. FairLens lets five agents reason freely, then audits their reasoning for bias in real time. YojanaSaathi keeps LLMs away from the decision entirely, because eligibility should be deterministic. Both come from the same principle: know exactly where intelligence belongs, and where it doesn't.
- π Β Building β multi-agent AI products with FastAPI, React and Google ADK
- π§ Β Exploring β evaluation loops, bias auditing, reliable LLM orchestration
- π― Β Open to β SDE / AI engineering roles and serious open-source work
- π¬ Β Reach me β sargarapremrajesh@gmail.com
|
Multi-agent hiring panel with real-time bias auditing Three expert agents evaluate a candidate in parallel. A fourth watches their reasoning and flags cognitive bias mid-sentence. A fifth synthesises a debiased verdict.
|
Open-source welfare-scheme engine for India One citizen profile becomes a ranked, auditable catalog of 1,000+ central and state schemes β with document-gap analysis and draft applications.
|
|
Explainable brain-tumour detection from MRI Deep-learning classification with Grad-CAM heatmaps, so a prediction can be reviewed instead of merely trusted.
|
End-to-end encrypted real-time messaging A from-scratch messaging app where the security model is the product, not a feature badge.
|
More builds: Software-MRI Β· DarkGuard-AI Β· CodeViz Β· Academic-Archive
| Focus | What I'm doing |
|---|---|
| π€ AI systems | Agent orchestration, evaluation loops, bias auditing |
| βοΈ Backend | FastAPI, Postgres RLS, SSE, rate limiting, failure modes |
| ποΈ System design | HLD/LLD, caching, queues, designing for scale |
| π DSA | LeetCode and Codeforces, consistently |
| π Open source | GSoC preparation and DevTools / ML contributions |



